CN115270634A - Counterfactual interpretation generation method and system suitable for autonomous air combat field - Google Patents
Counterfactual interpretation generation method and system suitable for autonomous air combat field Download PDFInfo
- Publication number
- CN115270634A CN115270634A CN202210930992.2A CN202210930992A CN115270634A CN 115270634 A CN115270634 A CN 115270634A CN 202210930992 A CN202210930992 A CN 202210930992A CN 115270634 A CN115270634 A CN 115270634A
- Authority
- CN
- China
- Prior art keywords
- counterfactual
- sample
- individuals
- population
- optimization
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 50
- 238000005457 optimization Methods 0.000 claims abstract description 142
- 238000012549 training Methods 0.000 claims abstract description 12
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 11
- 230000002068 genetic effect Effects 0.000 claims abstract description 9
- 230000006870 function Effects 0.000 claims description 29
- 230000001364 causal effect Effects 0.000 claims description 26
- 238000013136 deep learning model Methods 0.000 claims description 6
- 238000010801 machine learning Methods 0.000 claims description 6
- 238000010276 construction Methods 0.000 claims description 5
- 238000012545 processing Methods 0.000 claims description 4
- 238000004364 calculation method Methods 0.000 claims description 3
- 230000035772 mutation Effects 0.000 claims 2
- 229910052731 fluorine Inorganic materials 0.000 claims 1
- 125000001153 fluoro group Chemical group F* 0.000 claims 1
- 238000002474 experimental method Methods 0.000 description 7
- 230000008569 process Effects 0.000 description 6
- 238000002679 ablation Methods 0.000 description 3
- 230000008859 change Effects 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 230000002159 abnormal effect Effects 0.000 description 2
- 235000000332 black box Nutrition 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 241000282326 Felis catus Species 0.000 description 1
- 235000009421 Myristica fragrans Nutrition 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 238000002790 cross-validation Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 239000001115 mace Substances 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 230000002787 reinforcement Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012163 sequencing technique Methods 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 238000004326 stimulated echo acquisition mode for imaging Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
- G06F2119/02—Reliability analysis or reliability optimisation; Failure analysis, e.g. worst case scenario performance, failure mode and effects analysis [FMEA]
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Computer Hardware Design (AREA)
- Geometry (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to a counterfactual interpretation generation method and a counterfactual interpretation generation system which are suitable for the field of autonomous air combat, and the counterfactual interpretation generation method comprises the following steps: acquiring a simulated air combat data set, carrying out low-dimensional manifold representation, and constructing a multi-objective optimization function of an optimal counterfactual sample according to optimal counterfactual properties; constructing a black box model to be explained; training a black box model to be explained by using a low-dimensional fact sample and a corresponding decision label; applying a third-generation non-dominated sorting genetic algorithm according to the low-dimensional fact sample, the trained black box model to be explained and the multi-objective optimization function to obtain an optimal counterfactual sample and further obtain an optimal counterfactual sample explanation text; and obtaining a decision strategy of the autonomous air combat in the black box model to be explained based on the optimal counterfactual sample explanation text. And generating an optimal counterfactual sample more accurately based on the low-dimensional fact sample and a more comprehensive multi-objective optimization function, and further obtaining an optimal counterfactual interpretation text.
Description
Technical Field
The invention relates to the field of counterfactual interpretation in the field of autonomous air combat, in particular to a counterfactual interpretation generation method and a counterfactual interpretation generation system suitable for the field of autonomous air combat.
Background
Autonomous Air Combat (AAC) refers to airplaneThe technology of autonomously performing battlefield sensing, decision and control by means of relevant equipment such as an airborne device and the like to execute air battle. Autonomous air combat based on reinforcement learning has achieved combat performance exceeding that of human pilots, but its black-box nature has become a bottleneck for human-computer interaction and landing applications. By using a counterfactual research method in the fields of economics and psychology, counterfactual interpretation becomes an important way for revealing an internal mechanism of a black box model and generating high-order semantic interpretation. The opposite fact, i.e. given the original sample x 0 Generating a model f as close as possible to x with the black box model f to be explained 0 And the counterfactual sample x with the different prediction label cf Comparison of x 0 And x cf The difference of the two can be known as the key characteristics for determining the local decision rule of the black box model. At present, a local post-hoc interpretation method based on counterfactual samples becomes an important way for revealing an internal mechanism of a black box decision model, but the existing method for generating counterfactual samples and counterfactual interpretations has the following disadvantages: (1) The modeling method for counterfactual generation is different from the modeling method for counterfactual generation, and common recognition is difficult, for example, part of methods directly carry out disturbance in an original data feature space, and other methods emphasize that manifold representation of data needs to be learned firstly to improve causal feasibility and interpretability of a generated sample, and for example, modeling diversity of the counterfactual generation problem can be regarded as a process of finding a shortest path, a multi-target optimization problem or a Markov decision process and the like; (2) The optimal counterfactual properties are not completely modeled, and most of the existing methods only relate to partial optimal counterfactual sample properties; (3) The user needs natural language interpretation of high-level semantics, the counterfactual sample is only an intermediate result, but the existing method mostly ignores how to span the semantic gap between the counterfactual sample and the text interpretation. In contrast, the invention provides a counterfactual interpretation and generation method and system suitable for the field of autonomous air combat.
Disclosure of Invention
The invention aims to provide a counterfactual interpretation and generation method and a counterfactual interpretation and generation system which are suitable for the field of autonomous air combat.
In order to achieve the purpose, the invention provides the following scheme:
a counterfactual interpretation generation method suitable for the field of autonomous air combat comprises the following steps:
acquiring a simulated air combat data set and carrying out low-dimensional manifold representation on the simulated air combat data set to obtain low-dimensional air combat data; the low-dimensional air combat data comprises a plurality of low-dimensional fact samples and decision labels corresponding to the low-dimensional fact samples;
constructing a multi-objective optimization function of the optimal counterfactual sample according to the optimal counterfactual properties;
constructing a black box model to be explained; the black box model to be explained is a machine learning model or a deep learning model;
training the black-box model to be interpreted using the low-dimensional fact samples and the corresponding decision labels, obtaining a trained black box model to be explained;
obtaining an optimal counterfactual sample by applying a third-generation non-dominated sorting genetic algorithm according to the low-dimensional actual sample, the trained black box model to be explained and the multi-objective optimization function of the optimal counterfactual sample;
interpreting the optimal counterfactual sample according to the characteristic values of the optimal counterfactual sample and the low-dimensional counterfactual sample and a preset interpretation text set to obtain an optimal counterfactual sample interpretation text;
and obtaining a decision strategy of the autonomous air combat in the black box model to be explained based on the optimal counterfactual sample explanation text.
The invention also provides a counterfactual interpretation and generation system applicable to the field of autonomous air combat, which comprises:
the low-dimensional air combat data acquisition module is used for acquiring a simulated air combat data set and carrying out low-dimensional manifold representation on the simulated air combat data set to obtain low-dimensional air combat data; the low-dimensional air combat data comprises a plurality of low-dimensional fact samples and decision labels corresponding to the low-dimensional fact samples;
the optimization target determining module is used for constructing a multi-target optimization function of the optimal counterfactual sample according to the optimal counterfactual properties;
the model construction module is used for constructing a black box model to be explained; the black box model to be explained is a machine learning model or a deep learning model;
the training module is used for training the black box model to be interpreted by using the low-dimensional fact sample and the corresponding decision label to obtain the trained black box model to be interpreted;
the optimal counterfactual sample acquisition module is used for applying a third-generation non-dominated sorting genetic algorithm according to the low-dimensional actual sample, the trained black box model to be explained and the multi-target optimization function of the optimal counterfactual sample to obtain an optimal counterfactual sample;
the interpretation text generation module is used for interpreting the optimal counterfactual sample according to the characteristic values of the optimal counterfactual sample and the low-dimensional counterfactual sample and a preset interpretation text set to obtain an optimal counterfactual sample interpretation text;
and the decision strategy acquisition module is used for obtaining the decision strategy of the autonomous air combat in the black box model to be explained based on the optimal counterfactual sample explanation text.
According to the specific embodiment provided by the invention, the invention discloses the following technical effects:
the invention relates to a counterfactual interpretation generation method and a counterfactual interpretation generation system which are suitable for the field of autonomous air combat, and the counterfactual interpretation generation method comprises the following steps: acquiring a simulated air combat data set, carrying out low-dimensional manifold representation, and constructing a multi-objective optimization function of an optimal counterfactual sample according to optimal counterfactual properties; constructing a black box model to be explained; training a black box model to be explained by using a low-dimensional fact sample and a corresponding decision label; obtaining an optimal counterfactual sample and further obtaining an optimal counterfactual sample explanation text by applying a third-generation non-dominated sorting genetic algorithm according to a low-dimensional factual sample, a trained black box model to be explained and a multi-objective optimization function; and obtaining a decision strategy of the autonomous air combat in the black box model to be explained based on the optimal counterfactual sample explanation text. And generating an optimal counterfactual sample more accurately based on the low-dimensional fact sample and a more comprehensive multi-objective optimization function, and further obtaining an optimal counterfactual interpretation text.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without creative efforts.
Fig. 1 is a flowchart of a counterfactual interpretation and generation method suitable for the field of autonomous air combat provided in embodiment 1 of the present invention;
fig. 2 is a framework diagram of a counterfactual interpretation and generation method suitable for the field of autonomous air combat provided in embodiment 1 of the present invention;
FIG. 3 is the representation of AACE in different embedding dimensions on the DCS-AtoA data set provided by embodiment 1 of the present invention;
FIG. 4 is the representation of AACE in different embedding dimensions on the DCS-AtoG data set provided by embodiment 1 of the present invention;
fig. 5 is a counterfactual interpretation generation and system block diagram applicable to the field of autonomous air combat provided in embodiment 2 of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The invention aims to provide a counterfactual interpretation and generation method and a counterfactual interpretation and generation system which are suitable for the field of autonomous air combat.
In order to make the aforementioned objects, features and advantages of the present invention more comprehensible, the present invention is described in detail with reference to the accompanying drawings and the detailed description thereof.
Example 1
As shown in fig. 1 and 2, the present embodiment provides a counterfactual interpretation and generation method suitable for the field of autonomous air combat, including:
s1: acquiring a simulated air combat data set and carrying out low-dimensional manifold representation on the simulated air combat data set to obtain low-dimensional air combat data; the low-dimensional air war data comprises a plurality of low-dimensional fact samples and decision labels corresponding to the low-dimensional fact samples.
The method proposed herein was validated and studied using two air war data sets DCS-AtoA and DCS-AtoG. DCS-AtoA is data simulating an empty-to-empty dog fight scene, decision labels are classified into two categories (0-no fire, 1-fire), DCS-AtoG is data simulating an empty-to-earth attack scene, and decision labels are classified into five categories (0-no motion, 1-search, 2-aiming, 3-attack, 4-detachment). The two data sets are table data and are acquired based on a DCS-World air combat simulator sold by a STEAM platform.
A low-dimensional manifold representation of the data is learned using dfencoder. The dfencoder is a self-encoder suitable for table data, and the core idea of the dfencoder is that continuous features, multi-class features and binary features are respectively trained into three potential variables, and data points with large construction errors are removed to reduce abnormal deviation:
wherein L is an encoder, L -1 Is a decoder, X con ,X cat ,X bin Respectively, the continuous feature, the multi-class feature and the binary feature. argmin dist () represents to perform low-dimensional manifold representation, which represents to minimize reconstruction error and is a loss function of the training self-encoder; step S1 can learn the original simulationA low-dimensional data manifold representation of air combat data.
S2: and constructing a multi-objective optimization function of the optimal counterfactual sample according to the optimal counterfactual properties.
The optimal counterfactual needs to have six properties of legality, proximity, sparsity, diversity, causality feasibility and manifold proximity. Firstly, step S2 carries out optimization in the low-dimensional manifold representation obtained by learning in step S1 to generate counterfactual samples, which is done because the proximity requires that the change of the manifold points corresponding to the fact samples in the counterfactual optimization process is as small as possible, and the small change can be ignored when the counterfactual samples are generated through a decoder, so that the counterfactual samples are similar to the fact samples as much as possible, the manifold proximity is ensured, and the abnormal points do not appear. This is also the value of the self-encoder of step S1. Then given a label of y 0 Of the original sample x 0 Decision strategy to be interpreted pi θ Generated counterfactual samples x cf With the desired counterfactual tag y cf Counterfactual properties other than manifold proximity can be modeled as optimization objectives as follows. Specifically, step S2 includes:
s21: and determining a first optimization sub-target according to the fact sample with the different decision label from the low-dimensional fact sample.
Legitimacy: legitimacy ensures that the counterfactual exemplar labels generated are different from the original exemplar labels to satisfy the counterfactual properties. Due to y 0 The decision label predicted by the model is usually a probability, and the probability is compared with a specific threshold value to be converted into a discrete fire control label, so that the cross entropy loss is used for modeling the legality loss. Specifically, the expression of the first optimization sub-target is as follows:
loss validity =-y cf log(f(x cf ))+(1-y cf )log(1-f(x cf ))
among them, loss validity Representing a first optimization sub-goal; x is the number of cf Representing counterfactual samples; y is cf A decision label representing a counterfactual sample; f (x) cf ) And representing the prediction label of the black box model to be explained on the counterfactual sample.
S22: and determining a second optimization sub-target according to the similarity between the characteristics of the counterfactual sample and the characteristics of the low-dimensional actual sample.
The approach is as follows: proximity requires that the generated counterfactual samples be as close as possible to the actual samples on the basis of satisfying legitimacy, to satisfy the requirement of "closest to the contrary".
For continuity features, normalization is performed using the mean absolute deviation, since the scales of different features are different.
Wherein Conloss proximity A second optimization sub-goal representing a continuous-type feature; m represents the number of consecutive features in the counterfactual sample,an ith continuous feature representing a counterfactual sample; x is the number of i An ith continuous feature representing a fact sample; mad i Representing the mean absolute deviation of the ith successive feature of the fact sample.
For discrete features, proximity need only compare if the feature values are the same.
Catloss priximity A second optimization sub-goal representing a discrete-type feature; if it isThen I (· | ·) =0, otherwise I (· | ·) =1; n is the number of discrete features;
the proximity optimization objective can be modeled as a whole:
loss proximity =Conloss proximity +Catloss proximity
therein, loss proximity Representing a second optimization sub-goal.
S23: determining a third optimization sub-target according to the number of changed features in the counterfactual sample.
Sparsity: sparsity means that fewer features are changed in the generated counterfactual samples, thereby ensuring readability of the interpretation. Experiments show that continuous features are changed in the process of generating counterfactual samples, so that the invention only measures the number of changed discrete features as the optimization target of sparsity.
The expression of the third optimization sub-target is:
therein, loss sparsity Representing a third optimization sub-objective; if it isThen I (· | ·) =0, otherwise I (· | ·) =1.
S24: determining a fourth optimization sub-target according to the difference between a plurality of counterfactual samples generated by one low-dimensional fact sample;
diversity: diversity is a measure of the difference between multiple counterfactual samples generated by one fact sample. Higher diversity means that more viable choices and more information interpretations are provided to the user. A straightforward way to compute the loss of diversity is to sum the distances of each counterfactual pair, but in order to keep the sparsity low at the same time, the present invention implements the diversity constraint using a Deterministic Point Process (DPP), computing the determinant of the distance matrix.
The expression of the fourth optimization sub-objective is as follows:
among them, loss diversity Representing a fourth optimization sub-goal; dist (x) cfa ,x cfb ) The distance between the two counterfactual samples is calculated.
S25: determining a fifth optimization sub-goal according to causal constraints between features of the counterfactual samples.
Causal feasibility: causal feasibility preserves causal constraints between features. For example, education does not decline with age. This causal relationship is described in terms of a Structural Causal Model (SCM). The structured causal model is a directed acyclic graph whose nodes represent features and edges represent causal relationships from cause to effect. One node e in the SCM may be determined by its parent: e = g (pa (e),. Epsilon.). Where pa (e) represents the parent of e, g is a function representing causal strength, and e is Gaussian noise. The method first calculates a counterfactual sample with causal feasibility:
wherein,the l exogenous variable representing a counterfactual sample; the exogenous variable is a feature that is different from a continuous feature and a discrete feature in the counterfactual sample; g (-) refers to a structural causal model, and after the structural causal model G (-) and exogenous characteristic variables are given, all endogenous characteristic variables are calculated in a breadth-first traversal mode, so that an anti-fact sample with causal feasibility is obtained.
namely, the expression of the fifth optimization sub-target is as follows:
therein, loss causality Representing a fifth optimization sub-goal; (ii) a I O 2 Representing a two-norm.
S26: taking the counterfactual samples with the specified counterfactual labels in the low-dimensional fact samples as sixth optimization sub-targets; the counterfactual label refers to a decision label corresponding to the counterfactual sample.
Loss of prototype: to accelerate counterfactual generation, we add the optimization goal of prototype loss in addition to the optimal counterfactual properties. A prototype is a representation of a sample with a specified counter-fact label in a fact sample. The proto can be calculated by the following formula:
wherein,z is a hidden spatial representation of the low-dimensional fact samples, an exponential kernel function defined on the distance measure D;a hidden spatial representation of the p-th K-neighbor sample that is a low-dimensional fact sample that possesses a counter-fact label; k represents the number of K neighbor samples; the corner mark knn denotes the K-nearest neighbor algorithm.
The prototype loss, i.e. the expression of the sixth optimization sub-objective, is:
loss proto =||proto-z cf || 2
therein, loss proto Representing a sixth optimization sub-goal; z is a radical of cf A hidden spatial representation representing low-dimensional counterfactual samples.
S27: and constructing the multi-objective optimization function according to the first optimization sub-goal, the second optimization sub-goal, the third optimization sub-goal, the fourth optimization sub-goal, the fifth optimization sub-goal and the sixth optimization sub-goal.
The final goal of counterfactual optimization is a multi-objective optimization task as follows:
s3: constructing a black box model to be explained; the black box model to be explained is a machine learning model or a deep learning model.
S4: and training the black box model to be interpreted by using the low-dimensional fact sample and the corresponding decision label to obtain the trained black box model to be interpreted.
And (3) fitting the air battle data set by using the LightGBM, wherein the LightGBM is fitted with the original characteristic space, and ensuring that the AUC (estimate index) of the model reaches over 0.95 by adopting a 5-fold cross validation method.
S5: and obtaining the optimal counterfactual sample by applying a third generation non-dominated sorting genetic algorithm NSGA-III according to the low-dimensional fact sample, the trained black box model to be explained and the multi-objective optimization function of the optimal counterfactual sample.
Wherein, step S5 specifically includes:
s51: randomly generating an initial population subject to a Gaussian distribution based on a number of the low-dimensional fact samples.
S52: and crossing and mutating the individuals in the initial population to generate an intermediate population.
S53: and acquiring a set of the initial population and the intermediate population to obtain a set population.
S54: and inputting each individual in the set population into the trained black box model to be explained to obtain a decision label corresponding to each individual and calculating a plurality of optimization target values by combining the multi-objective optimization function.
In step S54, the individuals (counterfactual samples) in the ensemble population are input into the trained black box model to be interpreted to obtain the decision labels corresponding to the counterfactual samples, so as to participate in the calculation of multi-objective optimization.
S55: and performing non-dominant sequencing on the set population according to the plurality of optimization target values to obtain a layered pareto frontier.
S56: and acquiring individuals with the same number as the individuals of the initial population from the layered pareto frontier to obtain a next generation population.
Considering that the individual numbers of the individual layers of the pareto front do not necessarily constitute the number of individuals required for the next generation population, i.e., the number of individuals of the initial population, an attempt is made to determine whether the number of individuals in the pareto front of the preceding h layers is equal to the number of individuals of the initial population.
Specifically, step S56 specifically includes:
judging whether the number of individuals in the pareto frontier of the front h layer is equal to the number of individuals of the initial population;
if so, taking the individuals in the front edge of the pareto of the first h layers as the individuals in the next generation population;
if the number of the individuals in the pareto frontier of the front h layer is less than the number of the individuals in the initial population, enabling h = h +1, and returning to the step of judging whether the number of the individuals in the pareto frontier of the front h layer is equal to the number of the individuals in the initial population;
if the number of the individuals in the front edge of the pareto in the front h-1 layer is larger than the number of the individuals in the front edge of the pareto in the front layer, taking the individuals in the front edge of the pareto in the front layer as part of individuals in the next generation of population, and calculating the number of the remaining individuals;
performing scaling processing on all the optimized target values corresponding to all the individuals in the set population and generating a plurality of reference points by utilizing a simplex method according to a scaling result;
calculating the frequency of the reference point in the ecological niche;
selecting the reference point with the occurrence frequency less than the preset frequency as a target reference point;
and selecting the individuals with the residual number of the individuals from the pareto frontier of the h layer according to the similarity between the individuals and the target reference point, and adding the individuals into the next generation population to obtain a complete next generation population.
S57: and judging whether the current iteration times are equal to the maximum iteration times or not.
If so, the next generation population is the optimal counterfactual sample set;
if not, the next generation population is made to be the initial population, and the step S52 is returned until the maximum iteration number is reached.
S6: and interpreting the optimal counterfactual sample according to the characteristic values of the optimal counterfactual sample and the low-dimensional factual sample and a preset interpretation text set to obtain an optimal counterfactual sample interpretation text.
The preset explanation text set is artificially established according to domain knowledge, such as text describing a typical situation in a close-range dog fighting scene. The triggering condition is that one or more feature values of the fact sample and the counterfactual sample satisfy a certain condition, and within a certain range, a specific feature change in the fact sample and the counterfactual sample may trigger generation of a corresponding interpretation text.
In step S6, the optimal counterfactual samples in the high-dimensional space may be interpreted based on the feature values of the optimal counterfactual samples in the high-dimensional space and the fact samples in the high-dimensional space.
S7: and obtaining a decision strategy of the autonomous air combat in the black box model to be explained based on the optimal counterfactual sample explanation text.
After the optimal counterfactual sample is obtained, the internal rule in the black box model can be found, and then the decision strategy of the autonomous air combat learned by the black box model can be known.
In this embodiment, the quality of the generated counterfactual sample has better performance in all six indexes, and the method is called as AACE. Experiments are carried out on DCS-AtoA and DCS-AtoG generated based on DCS-world simulation, and the quality of counterfactual samples is measured by adopting the following indexes:
(1) Legal rate (valRatio): i.e. the ratio of legal counterfactual samples to total counterfactual samples.
(2) Proximity (meanProx): i.e., the mean of the generated counterfactual samples and the distance of the fact samples, is defined the same as the proximity optimization objective.
(3) Diversity (diversity): i.e. a set of counterfactual samples generated for one fact sample, which is the average of the distances between two.
(4) Ratio of causal feasibility (causalRatio): i.e. the ratio of the samples satisfying the causal constraint in the generated counterfactual samples to the total counterfactual samples
(5) IM1: measure the manifold closeness of the counterfactual samples, defined as follows:
wherein AE is cf For self-coders (dimensionality reduction), AE, trained on counterfactual classes (not counterfactual samples) ori For a self-encoder that trains on the fact class, e is a nonzero fractional amount.
(6) IM2: the interpretability of a counterfactual sample is measured and defined as follows:
wherein, (5) and (6) are abbreviations of interpretive metric, i.e. two interpretability indexes, for measuring the manifold proximity of counterfactual samples. The foregoing properties lack manifold proximity and have been added. The formula was applied in the process of evaluating counterfactual samples, see table 1, table 2.
Wherein AE cf For an auto-encoder trained on counterfactual classes (not counterfactual samples), AE full For a self-encoder trained on all classes (i.e., the entire data set), e is a nonzero fractional amount.
Among the six indexes, the larger the legal rate, diversity and causal feasibility are, the better the two indexes are; the closer the IM1, IM2, the better.
A partial counterfactual generation method of nearly three years is selected as baseline, and compared with AACE, the effects on DCS-AtoA and DCS-AtoG are shown in the following table:
TABLE 1 Performance of counterfactual sample Generation on DCS-AtoA datasets
TABLE 2 Performance of counterfactual sample Generation on DCS-AtoG datasets
Analysis of the above table yields:
(1) All methods can generate 100% legal counter-fact samples. AACE exceeded all other baselines in proximity, causal feasibility, IM1 and IM2 indices, reaching the second highest performance in diversity.
(2) By penalizing even more against infeasible counter-facts, the causal feasibility of AACE is superior to ProCE and MACE, and the latter two also add causal constraints. It is reasonable to consider error propagation in a structured causal model.
To explore the contribution factors of AACE to low IM1 and IM2, the method of this example was further conducted to ablation experiments, which remove the self-encoder and prototype losses, respectively, and the results are shown in tables 3 and 4 below:
TABLE 3 ablation experiments of AACE on DCS-AtoA
TABLE 4 ablation experiments of AACE on DCS-AtoA
Studies found that IM1 and IM2 increased slightly on both data sets without prototype guidance. However, whether or not optimization is performed on a dense data manifold representation has a greater impact on both metrics. Based on the above comparison it is strongly suggested to add an auto-encoder or a variant thereof as an optimized device for counterfactual generation.
To verify the robustness of the self-encoder, the behavior of the AACE in different embedding dimensions is explored, see fig. 3 and 4.
The other indicators on DCS-AtoA other than IM2 are stable when the embedding size is from 4 to 7. For the AtoG dataset, all the indicators were stable at different embedding scales. The above experiments demonstrate the overall stability of the autoencoder used by the AACE.
And finally, generating an explanation text by taking the unfired label of the air-to-air task DCS-AtoA as an example. According to expert knowledge, an empty-to-empty dog fight most easily occurs within a distance of 500-1000 meters, which is an effective attack range for most aerial cannons. In addition, if the pilot can enter the range of the tail of the enemy aircraft, a better attack attitude can be filed, and the departure angle of the aircraft is generally within 30 degrees. Accordingly, an interpretation set is developedAnd flip-flopIs (AOA and dist are both the characteristics in DCS-AtoA, respectively representing the departure angle and the distance of the friend or foe):
for the following example, the explanation of the counter-fact is: if the distance is within 1000 meters, the aircraft will be on fire.
TABLE 5A factual-counterfactual sample pair for generating interpretations in the DCS-AtoA task
The quality of counterfactual generation and the effectiveness of text interpretation generation are fully verified step by step through a plurality of experiments. From the above experimental results, the solution proposed by the present invention is novel, reliable and effective.
Example 2
As shown in fig. 5, the present embodiment provides a counterfactual interpretation and generation system suitable for the field of autonomous air combat, including:
the low-dimensional air combat data acquisition module T1 is used for acquiring a simulated air combat data set and performing low-dimensional manifold representation on the simulated air combat data set to obtain low-dimensional air combat data; the low-dimensional air combat data comprises a plurality of low-dimensional fact samples and decision labels corresponding to the low-dimensional fact samples;
the optimization target determining module T2 is used for constructing a multi-target optimization function of the optimal counterfactual sample according to the optimal counterfactual properties;
the optimization target determining module T2 specifically includes:
a first optimization target determining submodule T21, configured to determine a first optimization sub-target according to that a decision label of the counterfactual sample is different from a decision label of the low-dimensional actual sample;
a second optimization goal determining submodule T22, configured to determine a second optimization sub-goal according to similarity between features of the counterfactual sample and features of the low-dimensional fact sample;
a third optimization target determination submodule T23, configured to determine a third optimization sub-target according to the number of changed features in the counterfactual sample;
a fourth optimization target determining submodule T24, configured to determine a fourth optimization sub-target according to differences between the counterfactual samples generated by one low-dimensional fact sample;
a fifth optimization target determination submodule T25, configured to determine a fifth optimization sub-target according to causal constraints among the features of the counterfactual samples;
a first optimization goal determining submodule T26, configured to use the counterfactual sample with the specified counterfactual label in the low-dimensional fact samples as a sixth optimization sub-goal; the counterfactual label refers to a decision label corresponding to the counterfactual sample;
and the optimization objective determination submodule T27 is used for constructing the multi-objective optimization function according to the first optimization sub-objective, the second optimization sub-objective, the third optimization sub-objective, the fourth optimization sub-objective, the fifth optimization sub-objective and the sixth optimization sub-objective.
The model construction module T3 is used for constructing a black box model to be explained; the black box model to be explained is a machine learning model or a deep learning model;
the training module T4 is used for training the black box model to be interpreted by using the low-dimensional fact sample and the corresponding decision label to obtain a trained black box model to be interpreted;
the optimal counterfactual sample obtaining module T5 is used for obtaining an optimal counterfactual sample by applying a third-generation non-dominated sorting genetic algorithm according to the low-dimensional factual sample, the trained black box model to be explained and a multi-target optimization function of the optimal counterfactual sample;
the optimal counterfactual sample obtaining module T5 specifically includes:
the initial population constructing submodule T51 is used for randomly generating an initial population obeying Gaussian distribution based on the plurality of low-dimensional fact samples;
a cross variation submodule T52, configured to cross and vary individuals in the initial population to generate an intermediate population;
the population merging submodule T53 is used for acquiring a set of the initial population and the intermediate population to obtain a set population;
the optimization target value calculation submodule T54 is used for inputting each individual in the set population into the trained black box model to be explained, obtaining a decision label corresponding to each individual and calculating a plurality of optimization target values by combining the multi-objective optimization function;
the non-dominant sorting submodule T55 is used for performing non-dominant sorting on the set population according to the plurality of optimized target values to obtain a layered pareto frontier;
a next generation population generation submodule T56, configured to obtain, from the layered pareto frontier, individuals whose number is the same as that of the initial population, and obtain a next generation population;
the next generation population generation submodule T56 specifically includes:
a judging unit T561, configured to judge whether the number of individuals in a pareto front edge of a front h layer is equal to the number of individuals of the initial population;
if so, taking the individuals in the front edge of the pareto of the first h layers as the individuals in the next generation population;
if the number of the individuals in the pareto frontier of the front h layer is less than the number of the individuals in the initial population, enabling h = h +1, and returning to the step of judging whether the number of the individuals in the pareto frontier of the front h layer is equal to the number of the individuals in the initial population;
if the number of the individuals in the front edge of the pareto in the front h-1 layer is larger than the number of the individuals in the front edge of the pareto in the front layer, taking the individuals in the front edge of the pareto in the front layer as part of individuals in the next generation of population, and calculating the number of the remaining individuals;
a reference point generating unit T562, configured to perform scaling processing on all the optimized target values corresponding to all the individuals in the ensemble population and generate a plurality of reference points by using a simplex method according to a scaling result;
a target reference point determining unit T563 configured to calculate the number of occurrences of the reference point in the niche, and select the reference point, the number of occurrences of which is less than a preset number, as a target reference point;
and the next generation population generating unit T564 is used for selecting the individuals with the residual number of individuals from the pareto frontier of the h-th layer according to the similarity between the individuals and the target reference point to be added into the next generation population to obtain a complete next generation population.
A judgment submodule T57, configured to judge whether the current iteration number is equal to the maximum iteration number;
if so, the next generation population is the optimal counterfactual sample set;
and if not, enabling the next generation population to be the initial population, and returning to the step of crossing and mutating the individuals in the initial population to generate an intermediate population until the maximum iteration number is reached.
An interpretation text generation module T6, configured to interpret the optimal counterfactual sample according to the feature values of the optimal counterfactual sample and the low-dimensional counterfactual sample and a preset interpretation text set, so as to obtain an optimal counterfactual sample interpretation text;
and the decision strategy obtaining module T7 is used for obtaining a decision strategy of the autonomous air combat in the black box model to be explained based on the optimal counterfactual sample explanation text.
The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. For the system disclosed by the embodiment, the description is relatively simple because the system corresponds to the method disclosed by the embodiment, and the relevant points can be referred to the method part for description.
The principle and the embodiment of the present invention are explained by applying specific examples, and the above description of the embodiments is only used to help understanding the method and the core idea of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, the specific embodiments and the application range may be changed. In view of the above, the present disclosure should not be construed as limiting the invention.
Claims (10)
1. A counterfactual interpretation and generation method suitable for the field of autonomous air combat is characterized by comprising the following steps:
acquiring a simulated air combat data set and carrying out low-dimensional manifold representation on the simulated air combat data set to obtain low-dimensional air combat data; the low-dimensional air combat data comprises a plurality of low-dimensional fact samples and decision labels corresponding to the low-dimensional fact samples;
constructing a multi-objective optimization function of the optimal counterfactual sample according to the optimal counterfactual properties;
constructing a black box model to be explained; the black box model to be explained is a machine learning model or a deep learning model;
training the black box model to be interpreted by using the low-dimensional fact sample and the corresponding decision label to obtain a trained black box model to be interpreted;
applying a third-generation non-dominated sorting genetic algorithm according to the low-dimensional fact sample, the trained black box model to be explained and the multi-objective optimization function of the optimal counterfactual sample to obtain the optimal counterfactual sample;
interpreting the optimal counterfactual sample according to the characteristic values of the optimal counterfactual sample and the low-dimensional counterfactual sample and a preset interpretation text set to obtain an optimal counterfactual sample interpretation text;
and obtaining a decision strategy of the autonomous air combat in the black box model to be explained based on the optimal counterfactual sample explanation text.
2. The method according to claim 1, wherein the multi-objective optimization function for constructing the optimal counterfactual sample according to the optimal counterfactual properties specifically comprises:
determining a first optimization sub-target according to the fact sample and the low-dimensional fact sample;
determining a second optimization sub-target according to the similarity between the characteristics of the counterfactual sample and the characteristics of the low-dimensional actual sample;
determining a third optimization sub-target according to the quantity of the changed features in the counterfactual sample;
determining a fourth optimization sub-target according to the difference between a plurality of counterfactual samples generated by one low-dimensional fact sample;
determining a fifth optimization sub-goal according to causal constraints among the features of the counterfactual samples;
taking the counterfactual samples with the specified counterfactual labels in the low-dimensional fact samples as sixth optimization sub-targets; the counterfactual label refers to a decision label corresponding to the counterfactual sample;
and constructing the multi-objective optimization function according to the first optimization sub-goal, the second optimization sub-goal, the third optimization sub-goal, the fourth optimization sub-goal, the fifth optimization sub-goal and the sixth optimization sub-goal.
3. The method of claim 2, wherein the first optimization sub-goal is expressed as:
loss validity =-y cf log(f(x cf ))+(1-y cf )log(1-f(x cf ))
therein, loss validity Representing a first optimization sub-goal; x is a radical of a fluorine atom cf Representing counterfactual samples; y is cf A decision label representing a counterfactual sample; f (x) cf ) A prediction label representing a counterfactual sample of the black box model to be explained;
the expression of the second optimization sub-target is as follows:
loss proximity =Conloss proximity +Catloss proximity
therein, loss proximity Representing a second optimization sub-goal; conloss proximity A second optimization sub-goal representing a continuous type feature; catlos proximity A second optimization sub-goal representing a discrete-type feature;m represents the number of consecutive features in the counterfactual sample,an ith continuous feature representing a counterfactual sample; x is the number of i An ith continuous feature representing a fact sample; mad i Mean absolute deviation of the ith continuous feature representing a fact sample;if it isThen I (· | ·) =0, otherwise I (· | ·) =1; n is the number of discrete features;
the expression of the third optimization sub-target is:
therein, loss sparsity To representA third optimization sub-objective; if it isThen I (· | ·) =0, otherwise I (· | ·) =1;
the expression of the fourth optimization sub-target is as follows:
therein, loss diversity Representing a fourth optimization sub-goal; dist (x) cfa ,x cfb ) The distance between the two counterfactual samples is calculated.
4. The method of claim 2, wherein the expression of the fifth optimization sub-target is:
therein, loss causality Representing a fifth optimization sub-goal;g (-) refers to a structural causal model; epsilon represents the Gaussian noise; | | non-woven hair 2 Represents a two-norm;the l exogenous variable representing a counterfactual sample; the exogenous variable is a feature that is different from a continuous feature and a discrete feature in the counterfactual sample;
the expression of the sixth optimization sub-target is as follows:
loss proto =||proto-z cf || 2
among them, loss proto Representing a sixth optimization sub-goal;for an exponential kernel function defined on the distance measure D, z is a hidden spatial representation of the low-dimensional fact samples;a hidden spatial representation of the p-th K-neighbor sample that is a low-dimensional fact sample that possesses a counter-fact label; k represents the number of K neighbor samples; the corner mark knn represents the K-nearest neighbor algorithm.
5. The method according to claim 2, wherein the obtaining of the optimal counterfactual sample by applying a third-generation non-dominated sorting genetic algorithm according to the low-dimensional fact sample, the trained black box model to be interpreted, and the multi-objective optimization function of the optimal counterfactual sample comprises:
randomly generating an initial population obeying Gaussian distribution based on a plurality of the low-dimensional fact samples;
crossing and mutating the individuals in the initial population to generate an intermediate population;
acquiring a set of the initial population and the intermediate population to obtain a set population;
inputting each individual in the set population into the trained black box model to be interpreted to obtain a decision label corresponding to each individual and calculating a plurality of optimization target values by combining the multi-objective optimization function;
performing non-dominated sorting on the set population according to the plurality of optimized target values to obtain a layered pareto frontier;
obtaining individuals with the same number as the individuals of the initial population from the layered pareto frontier to obtain a next generation population;
judging whether the current iteration times are equal to the maximum iteration times or not;
if yes, the next generation population is the optimal counterfactual sample set;
and if not, enabling the next generation population to be the initial population, and returning to the step of crossing and mutating the individuals in the initial population to generate an intermediate population until the maximum iteration number is reached.
6. The method according to claim 5, wherein the obtaining of the same number of individuals as the initial population from the layered pareto frontier to obtain a next generation population comprises:
judging whether the number of individuals in the pareto frontier of the front h layer is equal to the number of individuals of the initial population;
if so, taking the individuals in the front edge of the pareto of the first h layers as the individuals in the next generation population;
if the sum is less than the preset threshold, h = h +1, and the step of judging whether the number of individuals in the pareto frontier of the front h layer is equal to the number of individuals of the initial population is returned;
if the population is larger than the threshold value, taking the individuals in the front edge of the pareto of the first h-1 layer as part of individuals in the next generation population, and calculating the number of the remaining individuals;
performing scaling processing on all the optimized target values corresponding to all the individuals in the set population and generating a plurality of reference points by utilizing a simplex method according to a scaling result;
calculating the frequency of the reference point in the ecological niche;
selecting the reference point with the occurrence frequency less than the preset frequency as a target reference point;
and selecting the individuals with the residual number of the individuals from the pareto frontier of the h layer according to the similarity between the individuals and the target reference point, and adding the individuals into the next generation population to obtain a complete next generation population.
7. A counterfactual interpretation generation system suitable for the field of autonomous air combat based on the method of any one of claims 1 to 6, comprising:
the low-dimensional air combat data acquisition module is used for acquiring a simulated air combat data set and carrying out low-dimensional manifold representation on the simulated air combat data set to obtain low-dimensional air combat data; the low-dimensional air combat data comprises a plurality of low-dimensional fact samples and decision labels corresponding to the low-dimensional fact samples;
the optimization target determining module is used for constructing a multi-target optimization function of the optimal counterfactual sample according to the optimal counterfactual property;
the model construction module is used for constructing a black box model to be explained; the black box model to be explained is a machine learning model or a deep learning model;
the training module is used for training the black box model to be interpreted by using the low-dimensional fact sample and the corresponding decision label to obtain the trained black box model to be interpreted;
the optimal counterfactual sample acquisition module is used for applying a third-generation non-dominated sorting genetic algorithm according to the low-dimensional actual sample, the trained black box model to be explained and the multi-target optimization function of the optimal counterfactual sample to obtain an optimal counterfactual sample;
the interpretation text generation module is used for interpreting the optimal counterfactual sample according to the characteristic values of the optimal counterfactual sample and the low-dimensional counterfactual sample and a preset interpretation text set to obtain an optimal counterfactual sample interpretation text;
and the decision strategy acquisition module is used for obtaining the decision strategy of the autonomous air combat in the black box model to be explained based on the optimal counterfactual sample explanation text.
8. The system of claim 7, wherein the optimization objective determination module specifically comprises:
the first optimization target determining submodule is used for determining a first optimization sub-target according to the fact sample counter decision label and the low-dimensional fact sample decision label;
a second optimization target determination submodule, configured to determine a second optimization sub-target according to similarity between features of the counterfactual sample and features of the low-dimensional fact sample;
a third optimization target determination sub-module for determining a third optimization sub-target according to the number of changed features in the counterfactual sample;
a fourth optimization target determining submodule, configured to determine a fourth optimization sub-target according to differences between the counterfactual samples generated by one low-dimensional fact sample;
a fifth optimization goal determining submodule for determining a fifth optimization sub-goal according to causal constraints among the features of the counter fact samples;
a first optimization goal determining submodule, configured to use the counterfactual sample with the specified counterfactual label in the low-dimensional fact sample as a sixth optimization sub-goal; the counterfactual label refers to a decision label corresponding to the counterfactual sample;
and the optimization goal determining submodule is used for constructing the multi-objective optimization function according to the first optimization sub-goal, the second optimization sub-goal, the third optimization sub-goal, the fourth optimization sub-goal, the fifth optimization sub-goal and the sixth optimization sub-goal.
9. The system according to claim 7, wherein the optimal counterfactual sample obtaining module specifically comprises:
the initial population construction submodule is used for randomly generating an initial population obeying Gaussian distribution based on a plurality of low-dimensional fact samples;
a cross mutation submodule for performing cross and mutation on the individuals in the initial population to generate an intermediate population;
the population merging submodule is used for acquiring a set of the initial population and the intermediate population to obtain a set population;
the optimization target value calculation submodule is used for inputting each individual in the set population into the trained black box model to be explained, obtaining a decision label corresponding to each individual and calculating a plurality of optimization target values by combining the multi-objective optimization function;
the non-dominance sorting submodule is used for carrying out non-dominance sorting on the set population according to the plurality of optimization target values to obtain a layered pareto front;
the next generation population generation submodule is used for acquiring individuals with the same number as the individuals of the initial population from the layered pareto frontier to obtain a next generation population;
the judgment submodule is used for judging whether the current iteration times are equal to the maximum iteration times or not;
if so, the next generation population is the optimal counterfactual sample set;
and if not, enabling the next generation population to be the initial population, and returning to the step of crossing and mutating the individuals in the initial population to generate an intermediate population until the maximum iteration number is reached.
10. The system of claim 9, wherein the next generation population generation submodule specifically comprises:
the judging unit is used for judging whether the number of individuals in the pareto frontier of the front h layer is equal to the number of individuals of the initial population;
if so, taking the individuals in the front edge of the pareto of the first h layers as the individuals in the next generation population;
if the number of the individuals in the pareto frontier of the front h layer is less than the number of the individuals in the initial population, enabling h = h +1, and returning to the step of judging whether the number of the individuals in the pareto frontier of the front h layer is equal to the number of the individuals in the initial population;
if the population is larger than the threshold value, taking the individuals in the front edge of the pareto of the first h-1 layer as part of individuals in the next generation population, and calculating the number of the remaining individuals;
a reference point generating unit, configured to perform scaling processing on all the optimized target values corresponding to all the individuals in the ensemble population and generate a plurality of reference points according to a scaling result by using a simplex method;
the target reference point determining unit is used for calculating the occurrence frequency of the reference point in the ecological niche and selecting the reference point with the occurrence frequency smaller than the preset frequency as a target reference point;
and the next generation population generating unit is used for selecting the individuals with the residual number of the individuals from the pareto frontier of the h layer according to the similarity between the individuals and the target reference point and adding the individuals into the next generation population to obtain a complete next generation population.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210930992.2A CN115270634B (en) | 2022-08-04 | 2022-08-04 | Counterfactual interpretation generation method and system suitable for autonomous air combat field |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210930992.2A CN115270634B (en) | 2022-08-04 | 2022-08-04 | Counterfactual interpretation generation method and system suitable for autonomous air combat field |
Publications (2)
Publication Number | Publication Date |
---|---|
CN115270634A true CN115270634A (en) | 2022-11-01 |
CN115270634B CN115270634B (en) | 2023-10-20 |
Family
ID=83749219
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210930992.2A Active CN115270634B (en) | 2022-08-04 | 2022-08-04 | Counterfactual interpretation generation method and system suitable for autonomous air combat field |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN115270634B (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116383731A (en) * | 2023-03-06 | 2023-07-04 | 南京航空航天大学 | Tactical maneuver identification method, tactical maneuver identification system, electronic equipment and storage medium |
CN116994687A (en) * | 2023-09-28 | 2023-11-03 | 之江实验室 | Clinical decision support model interpretation system based on inverse fact comparison |
CN117237096A (en) * | 2023-09-28 | 2023-12-15 | 大连理工大学 | Interpretable machine learning driven credit wind control system |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111240353A (en) * | 2020-01-07 | 2020-06-05 | 南京航空航天大学 | Unmanned aerial vehicle collaborative air combat decision method based on genetic fuzzy tree |
US20200293931A1 (en) * | 2019-03-15 | 2020-09-17 | Cognitive Scale, Inc. | Augmented Intelligence System Explainability Generation Engine |
CN112085252A (en) * | 2020-08-03 | 2020-12-15 | 清华大学 | Counterfactual prediction method about set type decision effect |
CN112541520A (en) * | 2019-09-20 | 2021-03-23 | 罗伯特·博世有限公司 | Apparatus and method for generating counterfactual data samples for neural networks |
US20210174253A1 (en) * | 2019-12-10 | 2021-06-10 | Triad National Security, Llc | Analysis of system log data using machine learning |
CN113269310A (en) * | 2021-06-03 | 2021-08-17 | 北京邮电大学 | Graph neural network interpretable method based on counterfactual |
US20220188645A1 (en) * | 2020-12-16 | 2022-06-16 | Oracle International Corporation | Using generative adversarial networks to construct realistic counterfactual explanations for machine learning models |
US11403538B1 (en) * | 2020-11-05 | 2022-08-02 | Arthur AI, Inc. | Methods and apparatus for generating fast counterfactual explanations for black-box models using reinforcement learning |
-
2022
- 2022-08-04 CN CN202210930992.2A patent/CN115270634B/en active Active
Patent Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20200293931A1 (en) * | 2019-03-15 | 2020-09-17 | Cognitive Scale, Inc. | Augmented Intelligence System Explainability Generation Engine |
CN112541520A (en) * | 2019-09-20 | 2021-03-23 | 罗伯特·博世有限公司 | Apparatus and method for generating counterfactual data samples for neural networks |
EP3796228A1 (en) * | 2019-09-20 | 2021-03-24 | Robert Bosch GmbH | Device and method for generating a counterfactual data sample for a neural network |
US20210174253A1 (en) * | 2019-12-10 | 2021-06-10 | Triad National Security, Llc | Analysis of system log data using machine learning |
CN111240353A (en) * | 2020-01-07 | 2020-06-05 | 南京航空航天大学 | Unmanned aerial vehicle collaborative air combat decision method based on genetic fuzzy tree |
CN112085252A (en) * | 2020-08-03 | 2020-12-15 | 清华大学 | Counterfactual prediction method about set type decision effect |
US11403538B1 (en) * | 2020-11-05 | 2022-08-02 | Arthur AI, Inc. | Methods and apparatus for generating fast counterfactual explanations for black-box models using reinforcement learning |
US20220188645A1 (en) * | 2020-12-16 | 2022-06-16 | Oracle International Corporation | Using generative adversarial networks to construct realistic counterfactual explanations for machine learning models |
CN113269310A (en) * | 2021-06-03 | 2021-08-17 | 北京邮电大学 | Graph neural network interpretable method based on counterfactual |
Non-Patent Citations (1)
Title |
---|
赵伟伟: "无人机集群编队及其避障控制关键技术研究", 中国博士学位论文数据库 工程科技Ⅱ辑 * |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116383731A (en) * | 2023-03-06 | 2023-07-04 | 南京航空航天大学 | Tactical maneuver identification method, tactical maneuver identification system, electronic equipment and storage medium |
CN116383731B (en) * | 2023-03-06 | 2023-11-14 | 南京航空航天大学 | Tactical maneuver identification method, tactical maneuver identification system, electronic equipment and storage medium |
CN116994687A (en) * | 2023-09-28 | 2023-11-03 | 之江实验室 | Clinical decision support model interpretation system based on inverse fact comparison |
CN117237096A (en) * | 2023-09-28 | 2023-12-15 | 大连理工大学 | Interpretable machine learning driven credit wind control system |
CN116994687B (en) * | 2023-09-28 | 2024-02-06 | 之江实验室 | Clinical decision support model interpretation system based on inverse fact comparison |
CN117237096B (en) * | 2023-09-28 | 2024-02-23 | 大连理工大学 | Interpretable machine learning driven credit wind control system |
Also Published As
Publication number | Publication date |
---|---|
CN115270634B (en) | 2023-10-20 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN115270634A (en) | Counterfactual interpretation generation method and system suitable for autonomous air combat field | |
Müller et al. | When does label smoothing help? | |
Bagheri et al. | Financial forecasting using ANFIS networks with quantum-behaved particle swarm optimization | |
Xu et al. | Composite quantile regression neural network with applications | |
Cao et al. | Class-specific soft voting based multiple extreme learning machines ensemble | |
Ranjan et al. | A computationally stable approach to Gaussian process interpolation of deterministic computer simulation data | |
Martens | Second-order optimization for neural networks | |
Denoyer et al. | Deep sequential neural network | |
CN111291556A (en) | Chinese entity relation extraction method based on character and word feature fusion of entity meaning item | |
Zhang et al. | Unsupervised graph embedding via adaptive graph learning | |
Fang et al. | Deep learning for turbulent channel flow | |
Han et al. | A Hybrid BPNN-GARF-SVR PredictionModel Based on EEMD for Ship Motion. | |
Bao et al. | Restricted Boltzmann Machine‐Assisted Estimation of Distribution Algorithm for Complex Problems | |
Tadayon et al. | Comparative analysis of the hidden markov model and lstm: A simulative approach | |
CN113626929A (en) | Multi-stage multi-topology ship traffic complexity measuring method and system | |
Xi et al. | Air Combat Maneuver Trajectory Prediction Model of Target Based on Chaotic Theory and IGA‐VNN | |
Osório et al. | Gaussian mixture models for affordance learning using bayesian networks | |
CN110866838A (en) | Network representation learning algorithm based on transition probability preprocessing | |
Da Cunha | Machine Learning for the Physical Sciences: Fundamentals and Prototyping with Julia | |
Halder et al. | An adaptive sampling algorithm for reduced‐order models using Isomap | |
Dominguez-Olmedo et al. | On data manifolds entailed by structural causal models | |
Li et al. | Dynamic Fuzzy Machine Learning | |
Abonyi et al. | System identification using delaunay tessellation of self-organizing maps | |
Narayanan et al. | Overview of Recent Advancements in Deep Learning and Artificial Intelligence | |
Kenesei et al. | Interpretability of computational intelligence-based regression models |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |